The Failed Demo
Historical product and cost clarification — October 11, 2026. The fintech sunset decision was around December 2025; the official pivot was around June 2026, customers received two years to transition and the app remains online. See the author’s clarification. The failed-demo scene is preserved. A monthly service price and one API call are different units: reliability, integrations, review, security, support and delivery obligations can add value beyond raw token cost. The historical external-company claims below are not proof that every software business is doomed. Jasper’s October 2022 announcement distinguishes $125M funding from its $1.5B valuation. For current work, see About.
I was demoing SimpleDirect to a potential customer last month.
Showing him the AI features. The chat. The document generation. The "smart" this, the "automated" that. Mid-demo energy, feeling pretty good about it.
Halfway through, he stops me.
"Wait - can't I just do this in ChatGPT?"
I sat there trying to come up with a reason he couldn't.
I couldn't.
Everything I was showing him - he could do for $20 a month. Or free.
I was charging $97.
My pitch died in my throat.
What I Was Actually Selling
Here's what I'd been building for years: a SaaS. SimpleDirect. Pricing tiers. $47, $97 - the whole playbook. The playbook I spent six years of my life learning.
We had AI-powered dashboards. Document generators. Financial tools. "Enterprise" features.
And then I actually looked at what we were building.
Our "AI features"? API calls. OpenAI. Anthropic. We'd take a prompt, send it to Claude, get a response back, and charge people $97 a month for the privilege.
The API call cost us about 3 cents.
We were going to charge $97.
A $97 monthly subscription and one $0.03 API call do not establish a comparable markup or service margin. Monthly usage and the full cost of delivery are missing.
And it's not even our technology. Same Claude. Same GPT. Same models anyone can access for $20 a month.
The Cope
I had all the cope ready.
"But we have integrations."
Zapier can help connect tools. The effort, reliability, permissions and maintenance needed for a particular production integration still have to be measured.
"But we have analytics."
SQL queries on data we were storing anyway.
"But we have workflows."
LangGraph. CrewAI. All free. All open source.
Every "feature" I thought was our moat was table stakes. I wasn't building IP. I was building configuration. And charging people $97 a month because they didn't know they could configure it themselves.
My entire business model was arbitraging their ignorance.
The Conversation
I had a conversation with a friend a few weeks later. He's running a startup. Paying Intercom about $40K a year for AI support.
I asked him: "Why don't you just build your own support bot?"
He said: "We're not a tech company. We don't have engineers for that."
He's a SaaS company. He has engineers. They're just building his product instead of replacing his $40K Intercom bill.
That excuse - "we can't"—it was true in 2020.
Now it's just not looking.
My claim about a weekend build is a personal report, not a measured guarantee that it meets another company’s production quality, escalation, privacy and support requirements.
The Math
Here's the math that broke my brain:
Intercom charges 99 cents per AI resolution. Plus seat fees. $29 to $139 per month. Mid-size company? $50K a year. Easy.
The actual API cost for that AI? About 2 cents per conversation.
99 cents vs 2 cents.
Same models. Same Claude. Same GPT.
Using only the stated $0.99 and $0.02 inputs gives 4,850% markup over that selected cost. But a resolved support issue and one model conversation are not necessarily equivalent units, and this excludes the other costs and responsibilities of the service.
The Graveyard
Grammarly was worth $13 billion. The AI grammar checker. 40 million users. $700 million in revenue.
This year, they acquired Coda. Then Superhuman.
Then they changed their name.
The company is now called Superhuman.
A $13 billion company looked at their core product—grammar checking, AI writing assistance - and decided they needed to run away from their own name.
Jasper announced a $125 million Series A at a $1.5 billion valuation in October 2022. The original paragraph conflated valuation with money raised; it did not establish a revenue decline caused by ChatGPT.
Builder.ai raised $445 million. Bankrupt. Turns out their "AI" was humans in offshore call centers.
Humane raised $241 million. Fire-sold.
Tome had 20 million users. Dead.
All in the last 12 months.
What Survives
Not all SaaS dies.
The ones that survive? They're not selling software. They're selling liability.
Payroll - Rippling, Gusto. One mistake means lawsuits. Regulatory hell. Angry employees not getting paid.
Compliance - Vanta, Drata. Same thing. Nobody trusts a weekend project with their SOC 2.
Banking - Stripe, Plaid. Legal nightmare if you get it wrong.
The moat isn't code. The moat is blame. You're paying for someone else to be responsible when it goes wrong.
But support software? CRM? Sales automation? Marketing tools? Project management?
For those products, evaluate actual outcomes, integration, accountability and delivery cost. The examples here do not establish that every product in these categories will fail.
The only question is how long the customers take to notice.
The Kill
I made the fintech sunset decision around December 2025. The official pivot followed around June 2026, with a two-year customer transition and the app still online.
SimpleDirect as SaaS. The pricing tiers. The subscriber goals. The whole thing.
I took the tools we built - Changelog, the AI chat, the stuff that was actually useful - and I made them free.
Not freemium. Free.
Not because I'm generous. Because I saw where this is going.
Charging people for an API wrapper in 2025 isn't a business. It's a toll booth on a road that's about to get demolished.
Still Processing
I still think about that demo.
The guy asking "can't I just do this in ChatGPT?"
The silence while I tried to find an answer.
That's the whole industry in one moment.
The vendor's moat is you not knowing what you could do yourself. That's the entire business model.
I was the vendor.
Still processing that.
The guidance and equity model described here belongs to this historical period. For my current work, use the About page linked in the dated note above.